Compressed vision information restoration based on cloud prior and local prior

Abstract

In wireless communication, compressed vision information may suffer from kinds of degradation, which dramatically influences the final visual quality. In this paper, a compressed vision information restoration method is proposed based on two explored vision priors: 1) the cloud prior and 2) the local prior. The cloud prior can be obtained from the nature images set in the cloud, and fields of experts is used to formulate the statistical character of the nature image contents as a high order Markov random field. The local prior is achieved from the degraded image itself, and K-SVD is adopted to model the sparse and redundant representation characters of nature images. These priors are effectively comprised in the proposed vision information restoration method. The relation between the quantization parameter and the optimal configuration of the prior models is further analyzed. In addition, an enhanced quantization constrained projection algorithm is proposed to refine the high frequency components. We extend this paper to compressed video restoration for H.264/AVC and the experiment results demonstrate that the proposed scheme can reproduce higher quality images compared with conventional H.264/AVC.